Hourly Temperature Forecast Revision Based on Convolutional Neural Network Algorithm on Observations at Alpine Meteorological Stations in Jinhua Area of China
Smart grid temperature forecast products are subjected to geomorphology and climatic conditions,and the temperature forecast results are prone to errors in mountainous areas with complex terrain.Reducing the errors by manually revising observations at some key points of an alpine meteorological observatory network had a large subjectivity and limited forecast accuracy,making it difficult to meet the demand for refined meteorological services.Neural network algorithms offer an objective computational approach to greatly improving forecast accuracy by learning from historical temperature data to regulate current data;however,this method was more commonly applied in plains with small hourly temperature fluctuations,and seldom used in mountainous areas where temperature fluctuations are large.In this paper,the Jinhua mountainous area of Zhejiang province,China was taken as research target.The intelligent grid temperature forecast product of Zhejiang province were calibrated with simultaneous hourly temperature observations collected at alpine meteorological observation stations by convolutional neural network(CNN)algorithm for proper revisions on hourly temperature prediction at key points/stations.(1)On an hourly scale,the root mean square error of temperatures at each station significantly decreased after the CNN revision,from 3 ℃-7 ℃ before the revision to 2 ℃-3 ℃ after the revision,satisfying expectation of accuracy.(2)On a monthly scale,temperature prediction accuracy improved notably as compared with product by the intelligent grid.After the revision,the accuracy of the monthly average temperature increased by 33.18％to 46.86％,with the highest accuracy in June.(3)The CNN model was more stable in revising mountain temperature forecasts than the manual revision method.It justified that two key indicators(average absolute error and 2 ℃ forecast accuracy)were close to or exceeded the concurrent acceptance check standards of Zhejiang Provincial Weather Forecast Quality Inspection Platform.This study improves the operational availability of temperature forecast products in alpine meteorological observatory in Jinhua area,which can provide data support for refined mountain weather services in mountainous areas.
NETL
NSTL
万方数据
